<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Ml on Suriya's site</title><link>https://subygan.com/tags/ml/</link><description>Recent content in Ml on Suriya's site</description><generator>Hugo</generator><language>en</language><copyright>This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.</copyright><item><title>LLM's can jump</title><link>https://subygan.com/essays/jump/</link><pubDate>Fri, 31 Jul 2026 00:00:00 +0000</pubDate><guid>https://subygan.com/essays/jump/</guid><description>&lt;p>I was reading &lt;a href="https://openreview.net/attachment?id=klU4737opt&amp;amp;name=originally_submitted_PDF">this paper&lt;/a>
,&lt;/p>
&lt;object data="https://openreview.net/attachment?id=klU4737opt&amp;amp;name=originally_submitted_PDF" type="application/pdf" width="100%" height="700px">
 &lt;p>It appears you don't have a PDF plugin for this browser.
 No biggie... you can &lt;a href="https://openreview.net/attachment?id=klU4737opt&amp;amp;name=originally_submitted_PDF">click here to download the PDF file.&lt;/a>&lt;/p></description></item><item><title>the genius who never learns</title><link>https://subygan.com/essays/continual-learning/</link><pubDate>Mon, 20 Jul 2026 00:00:00 +0000</pubDate><guid>https://subygan.com/essays/continual-learning/</guid><description>&lt;p>an agent that learns and optimizes will outperform a frozen 1.8 trillion parameter model.&lt;/p>
&lt;p>i&amp;rsquo;m not saying a 7B model is smarter than a frontier model. on a zero-context task the big models win. what i&amp;rsquo;m saying is that the moment you fix a task, a customer issue, a codebase, a distribution of tickets, and you let one system accumulate the corrections while the other one wakes up with amnesia every morning, the one that accumulates wins. and most valuable work is a fixed task with a customer and a distribution of tickets.&lt;/p></description></item></channel></rss>